diff --git a/Makefile b/Makefile index 7471ca92f..b621ff5e5 100644 --- a/Makefile +++ b/Makefile @@ -498,7 +498,7 @@ check: test .PHONY: install $(TAGS_PREREQ) install: $(wildcard *.go) - $(GO) install -v -tags '$(TAGS)' -ldflags '-s -w $(LDFLAGS)' + $(GO) install -v -tags '$(TAGS)' -ldflags ' $(LDFLAGS)' .PHONY: build build: frontend backend @@ -514,7 +514,7 @@ generate: $(TAGS_PREREQ) CC= GOOS= GOARCH= $(GO) generate -mod=vendor -tags '$(TAGS)' $(GO_PACKAGES) $(EXECUTABLE): $(GO_SOURCES) $(TAGS_PREREQ) - $(GO) build -mod=vendor $(GOFLAGS) $(EXTRA_GOFLAGS) -tags '$(TAGS)' -ldflags '-s -w $(LDFLAGS)' -o $@ + $(GO) build -mod=vendor $(GOFLAGS) $(EXTRA_GOFLAGS) -tags '$(TAGS)' -ldflags ' $(LDFLAGS)' -o $@ .PHONY: release release: frontend generate release-windows release-linux release-darwin release-copy release-compress release-sources release-check diff --git a/custom/conf/app.ini.sample b/custom/conf/app.ini.sample index 9fd00a763..9d6689493 100755 --- a/custom/conf/app.ini.sample +++ b/custom/conf/app.ini.sample @@ -1096,6 +1096,7 @@ LOCATION = cn-south-222 BASE_PATH = attachment/ [modelarts] +ORGANIZATION = modelarts ENDPOINT = https://modelarts.cn-south-222.ai.pcl.cn PROJECT_ID = edfccf24aace4e17a56da6bcbb55a5aa PROJECT_NAME = cn-south-222_test diff --git a/models/attachment.go b/models/attachment.go index 684a38b21..db486e8f3 100755 --- a/models/attachment.go +++ b/models/attachment.go @@ -429,7 +429,7 @@ func GetAllUserAttachments(userID int64) ([]*AttachmentUsername, error) { func getModelArtsUserAttachments(e Engine, userID int64) ([]*AttachmentUsername, error) { attachments := make([]*AttachmentUsername, 0, 10) if err := e.Table("attachment").Join("LEFT", "`user`", "attachment.uploader_id "+ - "= `user`.id").Where("attachment.type = ? and (uploader_id= ? or is_private = ?)", TypeCloudBrainTwo, userID, false).Find(&attachments); err != nil { + "= `user`.id").Where("attachment.type = ? and (uploader_id= ? or is_private = ?)", TypeCloudBrainNotebook, userID, false).Find(&attachments); err != nil { return nil, err } return attachments, nil diff --git a/models/cloudbrain.go b/models/cloudbrain.go index 4b2bec8e6..830eb16d5 100755 --- a/models/cloudbrain.go +++ b/models/cloudbrain.go @@ -5,6 +5,7 @@ import ( "fmt" "strings" "time" + "xorm.io/builder" "xorm.io/xorm" @@ -63,6 +64,10 @@ type Cloudbrain struct { CanDel bool `xorm:"-"` Type int `xorm:"INDEX DEFAULT 0"` + VersionID int64 `xorm:"INDEX DEFAULT 0"` + VersionName string + Uuid string + User *User `xorm:"-"` Repo *Repository `xorm:"-"` } @@ -530,7 +535,260 @@ type NotebookDelResult struct { InstanceID string `json:"instance_id"` } -func Cloudbrains(opts *CloudbrainsOptions) ([]*CloudbrainInfo, int64, error) { +type CreateTrainJobParams struct { + JobName string `json:"job_name"` + Description string `json:"job_desc"` + Config Config `json:"config"` + WorkspaceID string `json:"workspace_id"` +} + +type Config struct { + WorkServerNum int `json:"worker_server_num"` + AppUrl string `json:"app_url"` //训练作业的代码目录 + BootFileUrl string `json:"boot_file_url"` //训练作业的代码启动文件,需要在代码目录下 + Parameter []Parameter `json:"parameter"` + DataUrl string `json:"data_url"` //训练作业需要的数据集OBS路径URL + //DatasetID string `json:"dataset_id"` + //DataVersionID string `json:"dataset_version_id"` + //DataSource []DataSource `json:"data_source"` + //SpecID int64 `json:"spec_id"` + EngineID int64 `json:"engine_id"` + //ModelID int64 `json:"model_id"` + TrainUrl string `json:"train_url"` //训练作业的输出文件OBS路径URL + LogUrl string `json:"log_url"` + //UserImageUrl string `json:"user_image_url"` + //UserCommand string `json:"user_command"` + CreateVersion bool `json:"create_version"` + //Volumes []Volumes `json:"volumes"` + Flavor Flavor `json:"flavor"` + PoolID string `json:"pool_id"` +} + +type CreateConfigParams struct { + ConfigName string `json:"config_name"` + Description string `json:"config_desc"` + WorkServerNum int `json:"worker_server_num"` + AppUrl string `json:"app_url"` //训练作业的代码目录 + BootFileUrl string `json:"boot_file_url"` //训练作业的代码启动文件,需要在代码目录下 + Parameter []Parameter `json:"parameter"` + DataUrl string `json:"data_url"` //训练作业需要的数据集OBS路径URL + //DatasetID string `json:"dataset_id"` + //DataVersionID string `json:"dataset_version_id"` + //DataSource []DataSource `json:"data_source"` + //SpecID int64 `json:"spec_id"` + EngineID int64 `json:"engine_id"` + //ModelID int64 `json:"model_id"` + TrainUrl string `json:"train_url"` //训练作业的输出文件OBS路径URL + LogUrl string `json:"log_url"` + //UserImageUrl string `json:"user_image_url"` + //UserCommand string `json:"user_command"` + //CreateVersion bool `json:"create_version"` + //Volumes []Volumes `json:"volumes"` + Flavor Flavor `json:"flavor"` + PoolID string `json:"pool_id"` +} + +type Parameter struct { + Label string `json:"label"` + Value string `json:"value"` +} + +type Parameters struct { + Parameter []Parameter `json:"parameter"` +} + +type DataSource struct { + DatasetID string `json:"dataset_id"` + DatasetVersion string `json:"dataset_version"` + Type string `json:"type"` + DataUrl string `json:"data_url"` +} + +type Volumes struct { + Nfs Nfs `json:"nfs"` + HostPath HostPath `json:"host_path"` +} + +type Nfs struct { + ID string `json:"id"` + SourcePath string `json:"src_path"` + DestPath string `json:"dest_path"` + ReadOnly bool `json:"read_only"` +} + +type HostPath struct { + SourcePath string `json:"src_path"` + DestPath string `json:"dest_path"` + ReadOnly bool `json:"read_only"` +} + +type Flavor struct { + Code string `json:"code"` +} + +type CreateTrainJobResult struct { + ErrorCode string `json:"error_code"` + ErrorMsg string `json:"error_msg"` + IsSuccess bool `json:"is_success"` + JobName string `json:"job_name"` + JobID int64 `json:"job_id"` + Status int `json:"status"` + CreateTime int64 `json:"create_time"` + VersionID int64 `json:"version_id"` + ResourceID string `json:"resource_id"` + VersionName string `json:"version_name"` +} + +type CreateTrainJobConfigResult struct { + ErrorCode string `json:"error_code"` + ErrorMsg string `json:"error_msg"` + IsSuccess bool `json:"is_success"` +} + +type GetResourceSpecsResult struct { + ErrorCode string `json:"error_code"` + ErrorMsg string `json:"error_msg"` + IsSuccess bool `json:"is_success"` + SpecTotalCount int `json:"spec_total_count"` + Specs []Specs `json:"specs"` +} + +type Specs struct { + Core string `json:"core"` + Cpu string `json:"cpu"` + IsNoResource bool `json:"no_resource"` + GpuType string `json:"gpu_type"` + SpecID int64 `json:"spec_id"` + GpuNum int `json:"gpu_num"` + SpecCode string `json:"spec_code"` + Storage string `json:"storage"` + MaxNum int `json:"max_num"` + UnitNum int `json:"unit_num"` + InterfaceType int `json:"interface_type"` +} + +type GetConfigListResult struct { + ErrorCode string `json:"error_code"` + ErrorMsg string `json:"error_msg"` + IsSuccess bool `json:"is_success"` + ConfigTotalCount int `json:"config_total_count"` + ParaConfigs []ParaConfig `json:"configs"` +} + +type ParaConfig struct { + ConfigName string `json:"config_name"` + ConfigDesc string `json:"config_desc"` + CreateTime int64 `json:"create_time"` + EngineType int `json:"engine_type"` + EngineName string `json:"engine_name"` + EngineId int64 `json:"engine_id"` + EngineVersion string `json:"engine_version"` + UserImageUrl string `json:"user_image_url"` + UserCommand string `json:"user_command"` + Result GetConfigResult +} + +type GetConfigResult struct { + ErrorCode string `json:"error_code"` + ErrorMsg string `json:"error_msg"` + IsSuccess bool `json:"is_success"` + ConfigName string `json:"config_name"` + Description string `json:"config_desc"` + WorkServerNum int `json:"worker_server_num"` + AppUrl string `json:"app_url"` //训练作业的代码目录 + BootFileUrl string `json:"boot_file_url"` //训练作业的代码启动文件,需要在代码目录下 + Parameter []Parameter `json:"parameter"` + DataUrl string `json:"data_url"` //训练作业需要的数据集OBS路径URL + //DatasetID string `json:"dataset_id"` + //DataVersionID string `json:"dataset_version_id"` + //DataSource []DataSource `json:"data_source"` + //SpecID int64 `json:"spec_id"` + EngineID int64 `json:"engine_id"` + //ModelID int64 `json:"model_id"` + TrainUrl string `json:"train_url"` //训练作业的输出文件OBS路径URL + LogUrl string `json:"log_url"` + //UserImageUrl string `json:"user_image_url"` + //UserCommand string `json:"user_command"` + //CreateVersion bool `json:"create_version"` + //Volumes []Volumes `json:"volumes"` + Flavor Flavor `json:"flavor"` + PoolID string `json:"pool_id"` +} + +type ErrorResult struct { + ErrorCode string `json:"error_code"` + ErrorMsg string `json:"error_message"` + IsSuccess bool `json:"is_success"` +} + +type GetTrainJobResult struct { + IsSuccess bool `json:"is_success"` + JobName string `json:"job_name"` + JobID int64 `json:"job_id"` + Description string `json:"job_desc"` + IntStatus int `json:"status"` + Status string + LongCreateTime int64 `json:"create_time"` + CreateTime string + Duration int64 `json:"duration"` //训练作业的运行时间,单位为毫秒 + VersionID int64 `json:"version_id"` + ResourceID string `json:"resource_id"` + VersionName string `json:"version_name"` + PreVersionID int64 `json:"pre_version_id"` + WorkServerNum int `json:"worker_server_num"` + AppUrl string `json:"app_url"` //训练作业的代码目录 + BootFileUrl string `json:"boot_file_url"` //训练作业的代码启动文件,需要在代码目录下 + Parameter []Parameter `json:"parameter"` + DataUrl string `json:"data_url"` //训练作业需要的数据集OBS路径URL + //DatasetID string `json:"dataset_id"` + //DataVersionID string `json:"dataset_version_id"` + //DataSource []DataSource `json:"data_source"` + //SpecID int64 `json:"spec_id"` + EngineID int64 `json:"engine_id"` + EngineName string `json:"engine_name"` + EngineVersion string `json:"engine_version"` + //ModelID int64 `json:"model_id"` + TrainUrl string `json:"train_url"` //训练作业的输出文件OBS路径URL + LogUrl string `json:"log_url"` + //UserImageUrl string `json:"user_image_url"` + //UserCommand string `json:"user_command"` + //Volumes []Volumes `json:"volumes"` + Flavor Flavor `json:"flavor"` + PoolID string `json:"pool_id"` + PoolName string `json:"pool_name"` + NasMountPath string `json:"nas_mount_path"` + NasShareAddr string `json:"nas_share_addr"` + DatasetName string +} + +type GetTrainJobLogResult struct { + ErrorCode string `json:"error_code"` + ErrorMsg string `json:"error_msg"` + IsSuccess bool `json:"is_success"` + Content string `json:"content"` + Lines int `json:"lines"` + StartLine string `json:"start_line"` + EndLine string `json:"end_line"` +} + +type GetTrainJobLogFileNamesResult struct { + ErrorCode string `json:"error_code"` + ErrorMsg string `json:"error_msg"` + IsSuccess bool `json:"is_success"` + LogFileList []string `json:"log_file_list"` +} + +type TrainJobResult struct { + ErrorCode string `json:"error_code"` + ErrorMsg string `json:"error_msg"` + IsSuccess bool `json:"is_success"` +} + +type LogFile struct { + Name string +} + +func Cloudbrains(opts *CloudbrainsOptions) ([]*Cloudbrain, int64, error) { sess := x.NewSession() defer sess.Close() diff --git a/models/file_chunk.go b/models/file_chunk.go index b849f0108..8decb7b44 100755 --- a/models/file_chunk.go +++ b/models/file_chunk.go @@ -14,7 +14,10 @@ const ( ) const ( - TypeCloudBrainOne = 0 + TypeCloudBrainOne = 0 + TypeCloudBrainNotebook = 1 + TypeCloudBrainTrainJob = 2 + TypeCloudBrainTwo = 1 ) diff --git a/models/repo.go b/models/repo.go index 7f4bfebba..744e2744f 100755 --- a/models/repo.go +++ b/models/repo.go @@ -6,13 +6,14 @@ package models import ( - "code.gitea.io/gitea/modules/blockchain" "context" "crypto/md5" "errors" "fmt" "html/template" + "code.gitea.io/gitea/modules/blockchain" + // Needed for jpeg support _ "image/jpeg" "image/png" @@ -171,11 +172,11 @@ type Repository struct { NumOpenIssues int `xorm:"-"` NumPulls int NumClosedPulls int - NumOpenPulls int `xorm:"-"` - NumMilestones int `xorm:"NOT NULL DEFAULT 0"` - NumClosedMilestones int `xorm:"NOT NULL DEFAULT 0"` - NumOpenMilestones int `xorm:"-"` - NumCommit int64 `xorm:"NOT NULL DEFAULT 0"` + NumOpenPulls int `xorm:"-"` + NumMilestones int `xorm:"NOT NULL DEFAULT 0"` + NumClosedMilestones int `xorm:"NOT NULL DEFAULT 0"` + NumOpenMilestones int `xorm:"-"` + NumCommit int64 `xorm:"NOT NULL DEFAULT 0"` IsPrivate bool `xorm:"INDEX"` IsEmpty bool `xorm:"INDEX"` @@ -215,8 +216,8 @@ type Repository struct { CreatedUnix timeutil.TimeStamp `xorm:"INDEX created"` UpdatedUnix timeutil.TimeStamp `xorm:"INDEX updated"` - Hot int64 `xorm:"-"` - Active int64 `xorm:"-"` + Hot int64 `xorm:"-"` + Active int64 `xorm:"-"` } // SanitizedOriginalURL returns a sanitized OriginalURL @@ -2464,7 +2465,7 @@ func (repo *Repository) IncreaseCloneCnt() { } func UpdateRepositoryCommitNum(repo *Repository) error { - if _,err := x.Exec("UPDATE `repository` SET num_commit = ? where id = ?", repo.NumCommit, repo.ID); err != nil { + if _, err := x.Exec("UPDATE `repository` SET num_commit = ? where id = ?", repo.NumCommit, repo.ID); err != nil { return err } diff --git a/modules/auth/modelarts.go b/modules/auth/modelarts.go index 0be3e3882..f2e5aeed5 100755 --- a/modules/auth/modelarts.go +++ b/modules/auth/modelarts.go @@ -14,3 +14,32 @@ type CreateModelArtsForm struct { func (f *CreateModelArtsForm) Validate(ctx *macaron.Context, errs binding.Errors) binding.Errors { return validate(errs, ctx.Data, f, ctx.Locale) } + +type CreateModelArtsNotebookForm struct { + JobName string `form:"job_name" binding:"Required"` + Attachment string `form:"attachment"` + Description string `form:"description"` +} + +func (f *CreateModelArtsNotebookForm) Validate(ctx *macaron.Context, errs binding.Errors) binding.Errors { + return validate(errs, ctx.Data, f, ctx.Locale) +} + +type CreateModelArtsTrainJobForm struct { + JobName string `form:"job_name" binding:"Required"` + Attachment string `form:"attachment" binding:"Required"` + BootFile string `form:"boot_file" binding:"Required"` + WorkServerNumber int `form:"work_server_number" binding:"Required"` + EngineID int `form:"engine_id" binding:"Required"` + PoolID string `form:"pool_id" binding:"Required"` + Flavor string `form:"flavor" binding:"Required"` + Params string `form:"run_para_list" binding:"Required"` + Description string `form:"description"` + IsSaveParam string `form:"is_save_para"` + ParameterTemplateName string `form:"parameter_template_name"` + PrameterDescription string `form:"parameter_description"` +} + +func (f *CreateModelArtsTrainJobForm) Validate(ctx *macaron.Context, errs binding.Errors) binding.Errors { + return validate(errs, ctx.Data, f, ctx.Locale) +} diff --git a/modules/cloudbrain/cloudbrain.go b/modules/cloudbrain/cloudbrain.go index 0de1db9a6..749b86720 100755 --- a/modules/cloudbrain/cloudbrain.go +++ b/modules/cloudbrain/cloudbrain.go @@ -123,7 +123,8 @@ func GenerateTask(ctx *context.Context, jobName, image, command, uuid, codePath, JobName: jobName, SubTaskName: SubTaskName, JobType: jobType, - Type: models.TypeCloudBrainOne, + Type: models.TypeCloudBrainOne, + Uuid: uuid, }) if err != nil { diff --git a/modules/modelarts/modelarts.go b/modules/modelarts/modelarts.go index edd9d5d6b..ed56f73ea 100755 --- a/modules/modelarts/modelarts.go +++ b/modules/modelarts/modelarts.go @@ -1,22 +1,53 @@ package modelarts import ( + "encoding/json" + "path" + "strconv" + "code.gitea.io/gitea/models" "code.gitea.io/gitea/modules/context" "code.gitea.io/gitea/modules/log" "code.gitea.io/gitea/modules/setting" "code.gitea.io/gitea/modules/storage" - "encoding/json" - "path" ) const ( + //notebook storageTypeOBS = "obs" autoStopDuration = 4 * 60 * 60 DataSetMountPath = "/home/ma-user/work" NotebookEnv = "Python3" NotebookType = "Ascend" + FlavorInfo = "Ascend: 1*Ascend 910 CPU: 24 核 96GiB (modelarts.kat1.xlarge)" + + //train-job + ResourcePools = "{\"resource_pool\":[{\"id\":\"pool1328035d\", \"value\":\"专属资源池\"}]}" + Engines = "{\"engine\":[{\"id\":1, \"value\":\"Ascend-Powered-Engine\"}]}" + EngineVersions = "{\"version\":[{\"id\":118,\"value\":\"MindSpore-1.0.0-c75-python3.7-euleros2.8-aarch64\"}," + + "{\"id\":119,\"value\":\"MindSpore-1.1.1-c76-python3.7-euleros2.8-aarch64\"}," + + "{\"id\":120,\"value\":\"MindSpore-1.1.1-c76-tr5-python3.7-euleros2.8-aarch64\"}," + + "{\"id\":117,\"value\":\"TF-1.15-c75-python3.7-euleros2.8-aarch64\"}" + + "]}" + // FlavorInfos = "{\"flavor\":[{\"code\":\"modelarts.bm.910.arm.public.2\",\"value\":\"Ascend : 2 * Ascend 910 CPU:48 核 512GiB\"}," + + // "{\"code\":\"modelarts.bm.910.arm.public.8\",\"value\":\"Ascend : 8 * Ascend 910 CPU:192 核 2048GiB\"}," + + // "{\"code\":\"modelarts.bm.910.arm.public.4\",\"value\":\"Ascend : 4 * Ascend 910 CPU:96 核 1024GiB\"}," + + // "{\"code\":\"modelarts.bm.910.arm.public.1\",\"value\":\"Ascend : 1 * Ascend 910 CPU:24 核 256GiB\"}" + + // "]}" + CodePath = "/code/" + OutputPath = "/output/" + LogPath = "/log/" + JobPath = "/job/" + OrderDesc = "desc" //向下查询 + OrderAsc = "asc" //向上查询 + Lines = 20 + TrainUrl = "train_url" + DataUrl = "data_url" + PerPage = 10 + + SortByCreateTime = "create_time" + ConfigTypeCustom = "custom" ) var ( @@ -24,6 +55,50 @@ var ( FlavorInfos *models.FlavorInfos ) +type GenerateTrainJobReq struct { + JobName string + Uuid string + Description string + CodeObsPath string + BootFile string + DataUrl string + TrainUrl string + FlavorCode string + LogUrl string + PoolID string + WorkServerNumber int + EngineID int64 + Parameters []models.Parameter +} + +type VersionInfo struct { + Version []struct { + ID int `json:"id"` + Value string `json:"value"` + } `json:"version"` +} + +type Flavor struct { + Info []struct { + Code string `json:"code"` + Value string `json:"value"` + } `json:"flavor"` +} + +type Engine struct { + Info []struct { + ID int `json:"id"` + Value string `json:"value"` + } `json:"engine"` +} + +type ResourcePool struct { + Info []struct { + ID string `json:"id"` + Value string `json:"value"` + } `json:"resource_pool"` +} + func GenerateTask(ctx *context.Context, jobName, uuid, description string) error { var dataActualPath string if uuid != "" { @@ -78,13 +153,15 @@ func GenerateTask(ctx *context.Context, jobName, uuid, description string) error } err = models.CreateCloudbrain(&models.Cloudbrain{ + Status: string(models.JobWaiting), UserID: ctx.User.ID, RepoID: ctx.Repo.Repository.ID, JobID: jobResult.ID, JobName: jobName, JobType: string(models.JobTypeDebug), - Type: models.TypeCloudBrainTwo, + Type: models.TypeCloudBrainNotebook, + Uuid: uuid, }) if err != nil { @@ -93,3 +170,103 @@ func GenerateTask(ctx *context.Context, jobName, uuid, description string) error return nil } + +func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) error { + jobResult, err := createTrainJob(models.CreateTrainJobParams{ + JobName: req.JobName, + Description: req.Description, + Config: models.Config{ + WorkServerNum: req.WorkServerNumber, + AppUrl: req.CodeObsPath, + BootFileUrl: req.BootFile, + DataUrl: req.DataUrl, + EngineID: req.EngineID, + TrainUrl: req.TrainUrl, + LogUrl: req.LogUrl, + PoolID: req.PoolID, + CreateVersion: true, + Flavor: models.Flavor{ + Code: req.FlavorCode, + }, + Parameter: req.Parameters, + }, + }) + if err != nil { + log.Error("CreateJob failed: %v", err.Error()) + return err + } + + err = models.CreateCloudbrain(&models.Cloudbrain{ + Status: TransTrainJobStatus(jobResult.Status), + UserID: ctx.User.ID, + RepoID: ctx.Repo.Repository.ID, + JobID: strconv.FormatInt(jobResult.JobID, 10), + JobName: req.JobName, + JobType: string(models.JobTypeDebug), + Type: models.TypeCloudBrainTrainJob, + VersionID: jobResult.VersionID, + VersionName: jobResult.VersionName, + Uuid: req.Uuid, + }) + + if err != nil { + log.Error("CreateCloudbrain(%s) failed:%v", req.JobName, err.Error()) + return err + } + + return nil +} + +func TransTrainJobStatus(status int) string { + switch status { + case 0: + return "UNKNOWN" + case 1: + return "INIT" + case 2: + return "IMAGE_CREATING" + case 3: + return "IMAGE_FAILED" + case 4: + return "SUBMIT_TRYING" + case 5: + return "SUBMIT_FAILED" + case 6: + return "DELETE_FAILED" + case 7: + return "WAITING" + case 8: + return "RUNNING" + case 9: + return "KILLING" + case 10: + return "COMPLETED" + case 11: + return "FAILED" + case 12: + return "KILLED" + case 13: + return "CANCELED" + case 14: + return "LOST" + case 15: + return "SCALING" + case 16: + return "SUBMIT_MODEL_FAILED" + case 17: + return "DEPLOY_SERVICE_FAILED" + case 18: + return "CHECK_INIT" + case 19: + return "CHECK_RUNNING" + case 20: + return "CHECK_RUNNING_COMPLETED" + case 21: + return "CHECK_FAILED" + + default: + return strconv.Itoa(status) + } + + return "" +} diff --git a/modules/modelarts/resty.go b/modules/modelarts/resty.go index f91be5e31..d17478c94 100755 --- a/modules/modelarts/resty.go +++ b/modules/modelarts/resty.go @@ -1,13 +1,14 @@ package modelarts import ( - "code.gitea.io/gitea/modules/log" "crypto/tls" "encoding/json" "fmt" "net/http" + "strconv" "code.gitea.io/gitea/models" + "code.gitea.io/gitea/modules/log" "code.gitea.io/gitea/modules/setting" "github.com/go-resty/resty/v2" ) @@ -23,6 +24,9 @@ const ( urlGetToken = "/v3/auth/tokens" urlNotebook = "/demanager/instances" + urlTrainJob = "/training-jobs" + urlResourceSpecs = "/job/resource-specs" + urlTrainJobConfig = "/training-job-configs" errorCodeExceedLimit = "ModelArts.0118" ) @@ -104,7 +108,7 @@ sendjob: Post(HOST + "/v1/" + setting.ProjectID + urlNotebook) if err != nil { - return nil, fmt.Errorf("resty create job: %s", err) + return nil, fmt.Errorf("resty create notebook: %s", err) } if res.StatusCode() == http.StatusUnauthorized && retry < 1 { @@ -121,11 +125,11 @@ sendjob: } if len(response.ErrorCode) != 0 { - log.Error("CreateJob failed(%s): %s", response.ErrorCode, response.ErrorMsg) + log.Error("createNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg) if response.ErrorCode == errorCodeExceedLimit { response.ErrorMsg = "所选规格使用数量已超过最大配额限制。" } - return &result, fmt.Errorf("CreateJob failed(%s): %s", response.ErrorCode, response.ErrorMsg) + return &result, fmt.Errorf("createNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg) } return &result, nil @@ -210,6 +214,45 @@ sendjob: return &result, nil } +func DelNotebook(jobID string) (*models.NotebookDelResult, error) { + checkSetting() + client := getRestyClient() + var result models.NotebookDelResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetHeader("Content-Type", "application/json"). + SetAuthToken(TOKEN). + SetResult(&result). + Delete(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID) + + if err != nil { + return &result, fmt.Errorf("resty DelJob: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + var response models.NotebookResult + err = json.Unmarshal(res.Body(), &response) + if err != nil { + log.Error("json.Unmarshal failed: %s", err.Error()) + return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error()) + } + + if len(response.ErrorCode) != 0 { + log.Error("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg) + return &result, fmt.Errorf("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg) + } + + return &result, nil +} + func DelJob(jobID string) (*models.NotebookDelResult, error) { checkSetting() client := getRestyClient() @@ -287,3 +330,441 @@ sendjob: return &result, nil } + +func createTrainJob(createJobParams models.CreateTrainJobParams) (*models.CreateTrainJobResult, error) { + checkSetting() + client := getRestyClient() + var result models.CreateTrainJobResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetHeader("Content-Type", "application/json"). + SetAuthToken(TOKEN). + SetBody(createJobParams). + SetResult(&result). + Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob) + + if err != nil { + return nil, fmt.Errorf("resty create train-job: %s", err) + } + + req, _ := json.Marshal(createJobParams) + log.Info("%s", req) + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("createTrainJob failed(%s): %s", result.ErrorCode, result.ErrorMsg) + return &result, fmt.Errorf("createTrainJob failed(%s): %s", result.ErrorCode, result.ErrorMsg) + } + + return &result, nil +} + +func GetResourceSpecs() (*models.GetResourceSpecsResult, error) { + checkSetting() + client := getRestyClient() + var result models.GetResourceSpecsResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetHeader("Content-Type", "application/json"). + SetAuthToken(TOKEN). + SetResult(&result). + Get(HOST + "/v1/" + setting.ProjectID + urlResourceSpecs) + + if err != nil { + return nil, fmt.Errorf("resty GetResourceSpecs: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("GetResourceSpecs failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf("GetResourceSpecs failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("GetResourceSpecs failed(%s): %s", result.ErrorCode, result.ErrorMsg) + return &result, fmt.Errorf("GetResourceSpecs failed(%s): %s", result.ErrorCode, result.ErrorMsg) + } + + return &result, nil +} + +func CreateTrainJobConfig(req models.CreateConfigParams) (*models.CreateTrainJobConfigResult, error) { + checkSetting() + client := getRestyClient() + var result models.CreateTrainJobConfigResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetHeader("Content-Type", "application/json"). + SetAuthToken(TOKEN). + SetBody(req). + SetResult(&result). + Post(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig) + + if err != nil { + return nil, fmt.Errorf("resty CreateTrainJobConfig: %s", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + //temp, _ := json.Marshal(req) + //log.Info("%s", temp) + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("CreateTrainJobConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf("CreateTrainJobConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("CreateTrainJobConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg) + return &result, fmt.Errorf("CreateTrainJobConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg) + } + + return &result, nil +} + +func GetConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) { + checkSetting() + client := getRestyClient() + var result models.GetConfigListResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetQueryParams(map[string]string{ + "per_page": strconv.Itoa(perPage), + "page": strconv.Itoa(page), + "sortBy": sortBy, + "order": order, + "search_content": searchContent, + "config_type": configType, + }). + SetAuthToken(TOKEN). + SetResult(&result). + Get(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig) + + if err != nil { + return nil, fmt.Errorf("resty GetConfigList: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("GetConfigList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf("获取参数配置列表失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("GetConfigList failed(%s): %s", result.ErrorCode, result.ErrorMsg) + return &result, fmt.Errorf("获取参数配置列表失败(%s): %s", result.ErrorCode, result.ErrorMsg) + } + + return &result, nil +} + +func GetParaConfig(configName, configType string) (models.GetConfigResult, error) { + checkSetting() + client := getRestyClient() + var result models.GetConfigResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetQueryParams(map[string]string{ + "config_type": configType, + }). + SetAuthToken(TOKEN). + SetResult(&result). + Get(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig + "/" + configName) + + if err != nil { + return result, fmt.Errorf("resty GetParaConfig: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("GetParaConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return result, fmt.Errorf("获取参数配置详情失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("GetParaConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg) + return result, fmt.Errorf("获取参数配置详情失败(%s): %s", result.ErrorCode, result.ErrorMsg) + } + + return result, nil +} + +func GetTrainJob(jobID, versionID string) (*models.GetTrainJobResult, error) { + checkSetting() + client := getRestyClient() + var result models.GetTrainJobResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetAuthToken(TOKEN). + SetResult(&result). + Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID) + + if err != nil { + return nil, fmt.Errorf("resty GetTrainJob: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("GetTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf("获取作业详情失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("GetTrainJob(%s) failed", jobID) + return &result, fmt.Errorf("获取作业详情失败") + } + + return &result, nil +} + +func GetTrainJobLog(jobID, versionID, baseLine, logFile, order string, lines int) (*models.GetTrainJobLogResult, error) { + checkSetting() + client := getRestyClient() + var result models.GetTrainJobLogResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetQueryParams(map[string]string{ + "base_line": baseLine, + "lines": strconv.Itoa(lines), + "log_file": logFile, + "order": order, + }). + SetAuthToken(TOKEN). + SetResult(&result). + Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/aom-log") + + if err != nil { + return nil, fmt.Errorf("resty GetTrainJobLog: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("GetTrainJobLog failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf("获取作业日志失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("GetTrainJobLog(%s) failed", jobID) + return &result, fmt.Errorf("获取作业日志失败:%s", result.ErrorMsg) + } + + return &result, nil +} + +func GetTrainJobLogFileNames(jobID, versionID string) (*models.GetTrainJobLogFileNamesResult, error) { + checkSetting() + client := getRestyClient() + var result models.GetTrainJobLogFileNamesResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetAuthToken(TOKEN). + SetResult(&result). + Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/log/file-names") + + if err != nil { + return nil, fmt.Errorf("resty GetTrainJobLogFileNames: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("GetTrainJobLogFileNames failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf("GetTrainJobLogFileNames failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("GetTrainJobLogFileNames(%s) failed", jobID) + return &result, fmt.Errorf("获取作业日志文件失败:%s", result.ErrorMsg) + } + + return &result, nil +} + +func DelTrainJob(jobID string) (*models.TrainJobResult, error) { + checkSetting() + client := getRestyClient() + var result models.TrainJobResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetAuthToken(TOKEN). + SetResult(&result). + Delete(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID) + + if err != nil { + return &result, fmt.Errorf("resty DelTrainJob: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("DelTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf("删除训练作业失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("DelTrainJob(%s) failed", jobID) + return &result, fmt.Errorf("删除训练作业失败:%s", result.ErrorMsg) + } + + return &result, nil +} + +func StopTrainJob(jobID, versionID string) (*models.TrainJobResult, error) { + checkSetting() + client := getRestyClient() + var result models.TrainJobResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetAuthToken(TOKEN). + SetResult(&result). + Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/stop") + + if err != nil { + return &result, fmt.Errorf("resty StopTrainJob: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("StopTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf("停止训练作业失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("StopTrainJob(%s) failed", jobID) + return &result, fmt.Errorf("停止训练作业失败:%s", result.ErrorMsg) + } + + return &result, nil +} diff --git a/modules/setting/setting.go b/modules/setting/setting.go index 509483223..2eae22cab 100755 --- a/modules/setting/setting.go +++ b/modules/setting/setting.go @@ -468,6 +468,7 @@ var ( Bucket string Location string BasePath string + CodePathPrefix string UserBasePath string //modelarts config @@ -478,6 +479,7 @@ var ( ModelArtsUsername string ModelArtsPassword string ModelArtsDomain string + AllowedOrg string ProfileID string PoolInfos string Flavor string @@ -1198,6 +1200,7 @@ func NewContext() { Bucket = sec.Key("BUCKET").MustString("testopendata") Location = sec.Key("LOCATION").MustString("cn-south-222") BasePath = sec.Key("BASE_PATH").MustString("attachment/") + CodePathPrefix = sec.Key("CODE_PATH_PREFIX").MustString("code/") UserBasePath = sec.Key("BASE_PATH_USER").MustString("users/") PROXYURL = sec.Key("PROXY_URL").MustString("") @@ -1209,6 +1212,7 @@ func NewContext() { ModelArtsUsername = sec.Key("USERNAME").MustString("") ModelArtsPassword = sec.Key("PASSWORD").MustString("") ModelArtsDomain = sec.Key("DOMAIN").MustString("cn-south-222") + AllowedOrg = sec.Key("ORGANIZATION").MustString("") ProfileID = sec.Key("PROFILE_ID").MustString("") PoolInfos = sec.Key("POOL_INFOS").MustString("") Flavor = sec.Key("FLAVOR").MustString("") diff --git a/options/locale/locale_zh-CN.ini b/options/locale/locale_zh-CN.ini index 6dc0d410c..9b5c0808a 100755 --- a/options/locale/locale_zh-CN.ini +++ b/options/locale/locale_zh-CN.ini @@ -777,6 +777,52 @@ cloudbrain_task=任务名称 cloudbrain_operate=操作 cloudbrain_status_createtime=状态/创建时间 +modelarts.notebook=调试作业 +modelarts.train_job=训练作业 +modelarts.train_job.new=新建作业 +modelarts.train_job.basic_info=基本信息 +modelarts.train_job.job_status=作业状态 +modelarts.train_job.job_name=作业名称 +modelarts.train_job.version=作业版本 +modelarts.train_job.start_time=开始时间 +modelarts.train_job.dura_time=持续时间 +modelarts.train_job.description=作业描述 +modelarts.train_job.parameter_setting=参数设置 +modelarts.train_job.parameter_setting_info=参数信息 +modelarts.train_job.fast_parameter_setting=一键式参数配置 +modelarts.train_job.fast_parameter_setting_config=如您已保存过参数配置,可单击 +modelarts.train_job.fast_parameter_setting_config_link=这里 +modelarts.train_job.frames=常用框架 +modelarts.train_job.algorithm_origin=算法来源 +modelarts.train_job.AI_driver=AI引擎 +modelarts.train_job.start_file=启动文件 +modelarts.train_job.boot_file_helper=启动文件是您程序执行的入口文件,必须是以.py结尾的文件。 +modelarts.train_job.dataset=数据集 +modelarts.train_job.run_parameter=运行参数 +modelarts.train_job.add_run_parameter=增加运行参数 +modelarts.train_job.parameter_name=参数名 +modelarts.train_job.parameter_value=参数值 +modelarts.train_job.resource_setting=资源设置 +modelarts.train_job.resource_setting_info=资源信息 +modelarts.train_job.resource_pool=资源池 +modelarts.train_job.resource_type=资源类型 +modelarts.train_job.standard=规格 +modelarts.train_job.NAS_address=NAS地址 +modelarts.train_job.NAS_mount_path=NAS挂载路径 +modelarts.train_job.query_whether_save_parameter=保存作业参数 +modelarts.train_job.save_helper=保存当前作业的配置参数,后续您可以使用已保存的配置参数快速创建训练作业。 +modelarts.train_job.common_frame=常用框架 +modelarts.train_job.amount_of_compute_node=计算节点个数 +modelarts.train_job.job_parameter_name=作业参数名称 +modelarts.train_job.parameter_description=作业参数描述 +modelarts.log=日志 +modelarts.version_manage=版本管理 +modelarts.back=返回 +modelarts.train_job_para_admin=作业参数管理 +modelarts.train_job_para.edit=编辑 +modelarts.train_job_para.connfirm=确定 + + template.items=模板选项 template.git_content=Git数据(默认分支) template.git_hooks=Git 钩子 diff --git a/public/self/test.js b/public/self/test.js new file mode 100644 index 000000000..2839c76ab --- /dev/null +++ b/public/self/test.js @@ -0,0 +1,28 @@ + +function displayDir(uuid){ + console.log('uuid 1=' + uuid); + + var html="\ + \ + \ + 数据集名称\ + 数据集类型\ + 数据集描述\ + 数据集创建者\ + "; + + for (var i=0;i<1;i++){ + var row = "\ + \ + "+uuid+"\ + " + uuid +"\ + 测试\ + 测试\ + 测试\ + "; + html=html+row; + } + + document.getElementById('dataset-files-table').innerHTML=html; + console.log('uuid 2=' + uuid); +} diff --git a/routers/api/v1/api.go b/routers/api/v1/api.go index c6f3ee0ac..9dd773c4d 100755 --- a/routers/api/v1/api.go +++ b/routers/api/v1/api.go @@ -853,7 +853,15 @@ func RegisterRoutes(m *macaron.Macaron) { m.Get("/:jobid", repo.GetCloudbrainTask) }, reqRepoReader(models.UnitTypeCloudBrain)) m.Group("/modelarts", func() { - m.Get("/:jobid", repo.GetModelArtsTask) + m.Group("/notebook", func() { + m.Get("/:jobid", repo.GetModelArtsNotebook) + }) + m.Group("/train-job", func() { + m.Group("/:jobid", func() { + m.Get("", repo.GetModelArtsTrainJob) + m.Get("/log", repo.TrainJobGetLog) + }) + }) }, reqRepoReader(models.UnitTypeCloudBrain)) }, repoAssignment()) }) diff --git a/routers/api/v1/repo/modelarts.go b/routers/api/v1/repo/modelarts.go index 52b83b446..1ec4fa919 100755 --- a/routers/api/v1/repo/modelarts.go +++ b/routers/api/v1/repo/modelarts.go @@ -11,9 +11,10 @@ import ( "code.gitea.io/gitea/modules/log" "code.gitea.io/gitea/modules/modelarts" "net/http" + "strconv" ) -func GetModelArtsTask(ctx *context.APIContext) { +func GetModelArtsNotebook(ctx *context.APIContext) { var ( err error ) @@ -43,3 +44,81 @@ func GetModelArtsTask(ctx *context.APIContext) { }) } + +func GetModelArtsTrainJob(ctx *context.APIContext) { + var ( + err error + ) + + jobID := ctx.Params(":jobid") + repoID := ctx.Repo.Repository.ID + job, err := models.GetRepoCloudBrainByJobID(repoID, jobID) + if err != nil { + ctx.NotFound(err) + return + } + result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(job.VersionID, 10)) + if err != nil { + ctx.NotFound(err) + return + } + + job.Status = modelarts.TransTrainJobStatus(result.IntStatus) + err = models.UpdateJob(job) + if err != nil { + log.Error("UpdateJob failed:", err) + } + + ctx.JSON(http.StatusOK, map[string]interface{}{ + "JobID": jobID, + "JobStatus": job.Status, + }) + +} + +func TrainJobGetLog(ctx *context.APIContext) { + var ( + err error + ) + + log.Info("test") + + var jobID = ctx.Params(":jobid") + var logFileName = ctx.Query("file_name") + var baseLine = ctx.Query("base_line") + var order = ctx.Query("order") + + if order != modelarts.OrderDesc && order != modelarts.OrderAsc { + log.Error("order(%s) check failed", order) + ctx.JSON(http.StatusBadRequest, map[string]interface{}{ + "err_msg": "order check failed", + }) + return + } + + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error()) + ctx.JSON(http.StatusInternalServerError, map[string]interface{}{ + "err_msg": "GetCloudbrainByJobID failed", + }) + return + } + + result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines) + if err != nil { + log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error()) + ctx.JSON(http.StatusInternalServerError, map[string]interface{}{ + "err_msg": "GetTrainJobLog failed", + }) + return + } + + ctx.JSON(http.StatusOK, map[string]interface{}{ + "JobID": jobID, + "StartLine": result.StartLine, + "EndLine": result.EndLine, + "Content": result.Content, + "Lines": result.Lines, + }) +} diff --git a/routers/repo/attachment.go b/routers/repo/attachment.go index 9468d37db..5b946f832 100755 --- a/routers/repo/attachment.go +++ b/routers/repo/attachment.go @@ -1000,7 +1000,7 @@ func queryDatasets(ctx *context.Context, attachs []*models.AttachmentUsername) { } func checkTypeCloudBrain(typeCloudBrain int) error { - if typeCloudBrain != models.TypeCloudBrainOne && typeCloudBrain != models.TypeCloudBrainTwo { + if typeCloudBrain != models.TypeCloudBrainOne && typeCloudBrain != models.TypeCloudBrainNotebook { log.Error("type error:", typeCloudBrain) return errors.New("type error") } diff --git a/routers/repo/modelarts.go b/routers/repo/modelarts.go index 080c36377..625e0f566 100755 --- a/routers/repo/modelarts.go +++ b/routers/repo/modelarts.go @@ -1,14 +1,23 @@ package repo import ( - "code.gitea.io/gitea/modules/modelarts" "encoding/json" "errors" - "github.com/unknwon/com" + "fmt" + "io" + "net/http" + "os" + "path" "strconv" "strings" "time" + "code.gitea.io/gitea/modules/git" + "code.gitea.io/gitea/modules/modelarts" + "code.gitea.io/gitea/modules/obs" + "code.gitea.io/gitea/modules/storage" + "github.com/unknwon/com" + "code.gitea.io/gitea/models" "code.gitea.io/gitea/modules/auth" "code.gitea.io/gitea/modules/base" @@ -18,9 +27,18 @@ import ( ) const ( + // tplModelArtsNotebookIndex base.TplName = "repo/modelarts/notebook/index" + tplModelArtsNotebookIndex base.TplName = "repo/modelarts/notebook/index" + tplModelArtsNotebookNew base.TplName = "repo/modelarts/notebook/new" + tplModelArtsNotebookShow base.TplName = "repo/modelarts/notebook/show" + tplModelArtsIndex base.TplName = "repo/modelarts/index" tplModelArtsNew base.TplName = "repo/modelarts/new" tplModelArtsShow base.TplName = "repo/modelarts/show" + + tplModelArtsTrainJobIndex base.TplName = "repo/modelarts/trainjob/index" + tplModelArtsTrainJobNew base.TplName = "repo/modelarts/trainjob/new" + tplModelArtsTrainJobShow base.TplName = "repo/modelarts/trainjob/show" ) // MustEnableDataset check if repository enable internal cb @@ -30,6 +48,7 @@ func MustEnableModelArts(ctx *context.Context) { return } } + func ModelArtsIndex(ctx *context.Context) { MustEnableModelArts(ctx) repo := ctx.Repo.Repository @@ -249,3 +268,770 @@ func ModelArtsDel(ctx *context.Context) { ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts") } + +func NotebookIndex(ctx *context.Context) { + MustEnableModelArts(ctx) + repo := ctx.Repo.Repository + page := ctx.QueryInt("page") + if page <= 0 { + page = 1 + } + + ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{ + ListOptions: models.ListOptions{ + Page: page, + PageSize: setting.UI.IssuePagingNum, + }, + RepoID: repo.ID, + Type: models.TypeCloudBrainNotebook, + }) + if err != nil { + ctx.ServerError("Cloudbrain", err) + return + } + + for i, task := range ciTasks { + if task.Status == string(models.JobRunning) { + ciTasks[i].CanDebug = true + } else { + ciTasks[i].CanDebug = false + } + } + + pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5) + pager.SetDefaultParams(ctx) + ctx.Data["Page"] = pager + + ctx.Data["PageIsNotebook"] = true + ctx.Data["Tasks"] = ciTasks + ctx.HTML(200, tplModelArtsNotebookIndex) +} + +func NotebookNew(ctx *context.Context) { + ctx.Data["PageIsNotebook"] = true + + t := time.Now() + var jobName = jobNamePrefixValid(cutString(ctx.User.Name, 5)) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:] + ctx.Data["job_name"] = jobName + + attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID) + if err != nil { + ctx.ServerError("GetAllUserAttachments failed:", err) + return + } + + ctx.Data["attachments"] = attachs + ctx.Data["dataset_path"] = modelarts.DataSetMountPath + ctx.Data["env"] = modelarts.NotebookEnv + ctx.Data["notebook_type"] = modelarts.NotebookType + if modelarts.FlavorInfos == nil { + json.Unmarshal([]byte(setting.FlavorInfos), &modelarts.FlavorInfos) + } + ctx.Data["flavors"] = modelarts.FlavorInfos.FlavorInfo + ctx.HTML(200, tplModelArtsNew) +} + +func NotebookCreate(ctx *context.Context, form auth.CreateModelArtsNotebookForm) { + ctx.Data["PageIsNotebook"] = true + jobName := form.JobName + uuid := form.Attachment + description := form.Description + + err := modelarts.GenerateTask(ctx, jobName, uuid, description) + if err != nil { + ctx.RenderWithErr(err.Error(), tplModelArtsNotebookNew, &form) + return + } + + ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/notebook") +} + +func NotebookShow(ctx *context.Context) { + ctx.Data["PageIsNotebook"] = true + + var jobID = ctx.Params(":jobid") + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + ctx.Data["error"] = err.Error() + ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil) + return + } + + result, err := modelarts.GetJob(jobID) + if err != nil { + ctx.Data["error"] = err.Error() + ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil) + return + } + + if result != nil { + task.Status = result.Status + err = models.UpdateJob(task) + if err != nil { + ctx.Data["error"] = err.Error() + ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil) + return + } + + createTime, _ := com.StrTo(result.CreationTimestamp).Int64() + result.CreateTime = time.Unix(int64(createTime/1000), 0).Format("2006-01-02 15:04:05") + endTime, _ := com.StrTo(result.LatestUpdateTimestamp).Int64() + result.LatestUpdateTime = time.Unix(int64(endTime/1000), 0).Format("2006-01-02 15:04:05") + result.QueuingInfo.BeginTime = time.Unix(int64(result.QueuingInfo.BeginTimestamp/1000), 0).Format("2006-01-02 15:04:05") + result.QueuingInfo.EndTime = time.Unix(int64(result.QueuingInfo.EndTimestamp/1000), 0).Format("2006-01-02 15:04:05") + } + + ctx.Data["task"] = task + ctx.Data["jobID"] = jobID + ctx.Data["result"] = result + ctx.HTML(200, tplModelArtsNotebookShow) +} + +func NotebookDebug(ctx *context.Context) { + var jobID = ctx.Params(":jobid") + _, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + ctx.ServerError("GetCloudbrainByJobID failed", err) + return + } + + result, err := modelarts.GetJob(jobID) + if err != nil { + ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil) + return + } + + res, err := modelarts.GetJobToken(jobID) + if err != nil { + ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil) + return + } + + urls := strings.Split(result.Spec.Annotations.Url, "/") + urlPrefix := result.Spec.Annotations.TargetDomain + for i, url := range urls { + if i > 2 { + urlPrefix += "/" + url + } + } + + debugUrl := urlPrefix + "?token=" + res.Token + ctx.Redirect(debugUrl) +} + +func NotebookStop(ctx *context.Context) { + var jobID = ctx.Params(":jobid") + log.Info(jobID) + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + ctx.ServerError("GetCloudbrainByJobID failed", err) + return + } + + if task.Status != string(models.JobRunning) { + log.Error("the job(%s) is not running", task.JobName) + ctx.ServerError("the job is not running", errors.New("the job is not running")) + return + } + + param := models.NotebookAction{ + Action: models.ActionStop, + } + res, err := modelarts.StopJob(jobID, param) + if err != nil { + log.Error("StopJob(%s) failed:%v", task.JobName, err.Error()) + ctx.ServerError("StopJob failed", err) + return + } + + task.Status = res.CurrentStatus + err = models.UpdateJob(task) + if err != nil { + ctx.ServerError("UpdateJob failed", err) + return + } + + ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/notebook") +} + +func NotebookDel(ctx *context.Context) { + var jobID = ctx.Params(":jobid") + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + ctx.ServerError("GetCloudbrainByJobID failed", err) + return + } + + if task.Status != string(models.JobStopped) { + log.Error("the job(%s) has not been stopped", task.JobName) + ctx.ServerError("the job has not been stopped", errors.New("the job has not been stopped")) + return + } + + _, err = modelarts.DelNotebook(jobID) + if err != nil { + log.Error("DelJob(%s) failed:%v", task.JobName, err.Error()) + ctx.ServerError("DelJob failed", err) + return + } + + err = models.DeleteJob(task) + if err != nil { + ctx.ServerError("DeleteJob failed", err) + return + } + + ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/notebook") +} + +func TrainJobIndex(ctx *context.Context) { + MustEnableModelArts(ctx) + + can, err := canUserCreateTrainJob(ctx.User.ID) + if err != nil { + ctx.ServerError("canUserCreateTrainJob", err) + return + } + + ctx.Data["CanCreate"] = can + + repo := ctx.Repo.Repository + page := ctx.QueryInt("page") + if page <= 0 { + page = 1 + } + + tasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{ + ListOptions: models.ListOptions{ + Page: page, + PageSize: setting.UI.IssuePagingNum, + }, + RepoID: repo.ID, + Type: models.TypeCloudBrainTrainJob, + }) + if err != nil { + ctx.ServerError("Cloudbrain", err) + return + } + + pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5) + pager.SetDefaultParams(ctx) + ctx.Data["Page"] = pager + + ctx.Data["PageIsTrainJob"] = true + ctx.Data["Tasks"] = tasks + ctx.HTML(200, tplModelArtsTrainJobIndex) +} + +func TrainJobNew(ctx *context.Context) { + ctx.Data["PageIsTrainJob"] = true + + can, err := canUserCreateTrainJob(ctx.User.ID) + if err != nil { + ctx.ServerError("canUserCreateTrainJob", err) + return + } + + if !can { + log.Error("the user can not create train-job") + ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job")) + return + } + + t := time.Now() + var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:] + ctx.Data["job_name"] = jobName + + attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID) + if err != nil { + ctx.ServerError("GetAllUserAttachments failed:", err) + return + } + ctx.Data["attachments"] = attachs + + var resourcePools modelarts.ResourcePool + if err = json.Unmarshal([]byte(modelarts.ResourcePools), &resourcePools); err != nil { + ctx.ServerError("json.Unmarshal failed:", err) + return + } + ctx.Data["resource_pools"] = resourcePools.Info + + var engines modelarts.Engine + if err = json.Unmarshal([]byte(modelarts.Engines), &engines); err != nil { + ctx.ServerError("json.Unmarshal failed:", err) + return + } + ctx.Data["engines"] = engines.Info + + var versionInfos modelarts.VersionInfo + if err = json.Unmarshal([]byte(modelarts.EngineVersions), &versionInfos); err != nil { + ctx.ServerError("json.Unmarshal failed:", err) + return + } + ctx.Data["engine_versions"] = versionInfos.Version + + var flavorInfos modelarts.Flavor + if err = json.Unmarshal([]byte(setting.FlavorInfos), &flavorInfos); err != nil { + ctx.ServerError("json.Unmarshal failed:", err) + return + } + ctx.Data["flavor_infos"] = flavorInfos.Info + + outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + ctx.Data["train_url"] = outputObsPath + + configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom) + if err != nil { + ctx.ServerError("getConfigList failed:", err) + return + } + + ctx.Data["config_list"] = configList.ParaConfigs + + ctx.HTML(200, tplModelArtsTrainJobNew) +} + +func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) { + ctx.Data["PageIsTrainJob"] = true + jobName := form.JobName + uuid := form.Attachment + description := form.Description + workServerNumber := form.WorkServerNumber + engineID := form.EngineID + bootFile := form.BootFile + flavorCode := form.Flavor + params := form.Params + poolID := form.PoolID + isSaveParam := form.IsSaveParam + repo := ctx.Repo.Repository + codeLocalPath := setting.JobPath + jobName + modelarts.CodePath + codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath + outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + "/" + + can, err := canUserCreateTrainJob(ctx.User.ID) + if err != nil { + ctx.ServerError("canUserCreateTrainJob", err) + return + } + + if !can { + log.Error("the user can not create train-job") + ctx.RenderWithErr("the user can not create train-job", tplModelArtsTrainJobNew, &form) + return + } + + //param check + if err := paramCheckCreateTrainJob(form); err != nil { + log.Error("paramCheckCreateTrainJob failed:(%v)", err) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form) + return + } + + if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{}); err != nil { + log.Error("Failed to clone repository: %s (%v)", repo.FullName(), err) + ctx.RenderWithErr("Failed to clone repository", tplModelArtsTrainJobNew, &form) + return + } + + //todo: upload code (send to file_server todo this work?) + if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath); err != nil { + log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err) + ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobNew, &form) + return + } + + if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath); err != nil { + log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err) + ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobNew, &form) + return + } + + if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil { + log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err) + ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobNew, &form) + return + } + + //todo: del local code? + + var parameters models.Parameters + param := make([]models.Parameter, 0) + param = append(param, models.Parameter{ + Label: modelarts.TrainUrl, + Value: outputObsPath, + }, models.Parameter{ + Label: modelarts.DataUrl, + Value: dataPath, + }) + if len(params) != 0 { + err := json.Unmarshal([]byte(params), ¶meters) + if err != nil { + log.Error("Failed to Unmarshal params: %s (%v)", params, err) + ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobNew, &form) + return + } + + for _, parameter := range parameters.Parameter { + if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl { + param = append(param, models.Parameter{ + Label: parameter.Label, + Value: parameter.Value, + }) + } + } + } + + //save param config + if isSaveParam == "on" { + if form.ParameterTemplateName == "" { + log.Error("ParameterTemplateName is empty") + ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobNew, &form) + return + } + + _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{ + ConfigName: form.ParameterTemplateName, + Description: form.PrameterDescription, + DataUrl: dataPath, + AppUrl: codeObsPath, + BootFileUrl: codeObsPath + bootFile, + TrainUrl: outputObsPath, + Flavor: models.Flavor{ + Code: flavorCode, + }, + WorkServerNum: workServerNumber, + EngineID: int64(engineID), + LogUrl: logObsPath, + PoolID: poolID, + Parameter: param, + }) + + if err != nil { + log.Error("Failed to CreateTrainJobConfig: %v", err) + ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobNew, &form) + return + } + } + + req := &modelarts.GenerateTrainJobReq{ + JobName: jobName, + DataUrl: dataPath, + Description: description, + CodeObsPath: codeObsPath, + BootFile: codeObsPath + bootFile, + TrainUrl: outputObsPath, + FlavorCode: flavorCode, + WorkServerNumber: workServerNumber, + EngineID: int64(engineID), + LogUrl: logObsPath, + PoolID: poolID, + Uuid: uuid, + Parameters: param, + } + + err = modelarts.GenerateTrainJob(ctx, req) + if err != nil { + log.Error("GenerateTrainJob failed:%v", err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form) + return + } + + ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job") +} + +// readDir reads the directory named by dirname and returns +// a list of directory entries sorted by filename. +func readDir(dirname string) ([]os.FileInfo, error) { + f, err := os.Open(dirname) + if err != nil { + return nil, err + } + + list, err := f.Readdir(100) + f.Close() + if err != nil { + //todo: can not upload empty folder + if err == io.EOF { + return nil, nil + } + return nil, err + } + + //sort.Slice(list, func(i, j int) bool { return list[i].Name() < list[j].Name() }) + return list, nil +} + +func uploadCodeToObs(codePath, jobName, parentDir string) error { + files, err := readDir(codePath) + if err != nil { + log.Error("readDir(%s) failed: %s", codePath, err.Error()) + return err + } + + for _, file := range files { + if file.IsDir() { + input := &obs.PutObjectInput{} + input.Bucket = setting.Bucket + input.Key = parentDir + file.Name() + "/" + _, err = storage.ObsCli.PutObject(input) + if err != nil { + log.Error("PutObject(%s) failed: %s", input.Key, err.Error()) + return err + } + + if err = uploadCodeToObs(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil { + log.Error("uploadCodeToObs(%s) failed: %s", file.Name(), err.Error()) + return err + } + } else { + input := &obs.PutFileInput{} + input.Bucket = setting.Bucket + input.Key = setting.CodePathPrefix + jobName + "/code/" + parentDir + file.Name() + input.SourceFile = codePath + file.Name() + _, err = storage.ObsCli.PutFile(input) + if err != nil { + log.Error("PutFile(%s) failed: %s", input.SourceFile, err.Error()) + return err + } + } + } + + return nil +} + +func obsMkdir(dir string) error { + input := &obs.PutObjectInput{} + input.Bucket = setting.Bucket + input.Key = dir + _, err := storage.ObsCli.PutObject(input) + if err != nil { + log.Error("PutObject(%s) failed: %s", input.Key, err.Error()) + return err + } + + return nil +} + +func paramCheckCreateTrainJob(form auth.CreateModelArtsTrainJobForm) error { + if !strings.HasSuffix(form.BootFile, ".py") { + log.Error("the boot file(%s) must be a python file", form.BootFile) + return errors.New("启动文件必须是python文件") + } + + if form.WorkServerNumber > 25 || form.WorkServerNumber < 1 { + log.Error("the WorkServerNumber(%d) must be in (1,25)", form.WorkServerNumber) + return errors.New("计算节点数必须在1-25之间") + } + + return nil +} + +func TrainJobShow(ctx *context.Context) { + ctx.Data["PageIsTrainJob"] = true + + var jobID = ctx.Params(":jobid") + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil) + return + } + + attach, err := models.GetAttachmentByUUID(task.Uuid) + if err != nil { + log.Error("GetAttachmentByUUID(%s) failed:%v", jobID, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil) + return + } + + result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(task.VersionID, 10)) + if err != nil { + log.Error("GetJob(%s) failed:%v", jobID, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil) + return + } + + if result != nil { + result.CreateTime = time.Unix(int64(result.LongCreateTime/1000), 0).Format("2006-01-02 15:04:05") + result.Status = modelarts.TransTrainJobStatus(result.IntStatus) + result.DatasetName = attach.Name + } + + resultLogFile, resultLog, err := trainJobGetLog(jobID) + if err != nil { + log.Error("trainJobGetLog(%s) failed:%v", jobID, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil) + return + } + + ctx.Data["log_file_name"] = resultLogFile.LogFileList[0] + ctx.Data["log"] = resultLog + ctx.Data["task"] = task + ctx.Data["jobID"] = jobID + ctx.Data["result"] = result + ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow) +} + +func TrainJobGetLog(ctx *context.Context) { + ctx.Data["PageIsTrainJob"] = true + + var jobID = ctx.Params(":jobid") + var logFileName = ctx.Query("file_name") + var baseLine = ctx.Query("base_line") + var order = ctx.Query("order") + + if order != modelarts.OrderDesc && order != modelarts.OrderAsc { + log.Error("order(%s) check failed", order) + ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow) + return + } + + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil) + return + } + + result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines) + if err != nil { + log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil) + return + } + + ctx.Data["log"] = result + //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow) +} + +func trainJobGetLog(jobID string) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) { + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error()) + return nil, nil, err + } + + resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10)) + if err != nil { + log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error()) + return nil, nil, err + } + + result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), "", resultLogFile.LogFileList[0], modelarts.OrderDesc, modelarts.Lines) + if err != nil { + log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error()) + return nil, nil, err + } + + return resultLogFile, result, err +} + +func TrainJobDel(ctx *context.Context) { + var jobID = ctx.Params(":jobid") + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil) + return + } + + _, err = modelarts.DelTrainJob(jobID) + if err != nil { + log.Error("DelTrainJob(%s) failed:%v", task.JobName, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil) + return + } + + err = models.DeleteJob(task) + if err != nil { + ctx.ServerError("DeleteJob failed", err) + return + } + + ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job") +} + +func TrainJobStop(ctx *context.Context) { + var jobID = ctx.Params(":jobid") + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil) + return + } + + _, err = modelarts.StopTrainJob(jobID, strconv.FormatInt(task.VersionID, 10)) + if err != nil { + log.Error("StopTrainJob(%s) failed:%v", task.JobName, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil) + return + } + + ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job") +} + +func canUserCreateTrainJob(uid int64) (bool, error) { + org, err := models.GetOrgByName(setting.AllowedOrg) + if err != nil { + log.Error("get allowed org failed: ", setting.AllowedOrg) + return false, err + } + + return org.IsOrgMember(uid) +} + +func TrainJobGetConfigList(ctx *context.Context) { + ctx.Data["PageIsTrainJob"] = true + + var jobID = ctx.Params(":jobid") + var logFileName = ctx.Query("file_name") + var baseLine = ctx.Query("base_line") + var order = ctx.Query("order") + + if order != modelarts.OrderDesc && order != modelarts.OrderAsc { + log.Error("order(%s) check failed", order) + ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow) + return + } + + task, err := models.GetCloudbrainByJobID(jobID) + if err != nil { + log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil) + return + } + + result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines) + if err != nil { + log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error()) + ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil) + return + } + + ctx.Data["log"] = result + //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow) +} + +func getConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) { + var result models.GetConfigListResult + + list, err := modelarts.GetConfigList(perPage, page, sortBy, order, searchContent, configType) + if err != nil { + log.Error("GetConfigList failed:", err) + return &result, err + } + + for _, config := range list.ParaConfigs { + paraConfig, err := modelarts.GetParaConfig(config.ConfigName, configType) + if err != nil { + log.Error("GetParaConfig failed:", err) + return &result, err + } + + config.Result = paraConfig + } + + return list, nil +} diff --git a/routers/routes/routes.go b/routers/routes/routes.go index 0a17c832c..71c112477 100755 --- a/routers/routes/routes.go +++ b/routers/routes/routes.go @@ -970,6 +970,31 @@ func RegisterRoutes(m *macaron.Macaron) { }) m.Get("/create", reqRepoCloudBrainWriter, repo.ModelArtsNew) m.Post("/create", reqRepoCloudBrainWriter, bindIgnErr(auth.CreateModelArtsForm{}), repo.ModelArtsCreate) + + m.Group("/notebook", func() { + m.Get("", reqRepoCloudBrainReader, repo.NotebookIndex) + m.Group("/:jobid", func() { + m.Get("", reqRepoCloudBrainReader, repo.NotebookShow) + m.Get("/debug", reqRepoCloudBrainReader, repo.NotebookDebug) + m.Post("/stop", reqRepoCloudBrainWriter, repo.NotebookStop) + m.Post("/del", reqRepoCloudBrainWriter, repo.NotebookDel) + }) + m.Get("/create", reqRepoCloudBrainWriter, repo.NotebookNew) + m.Post("/create", reqRepoCloudBrainWriter, bindIgnErr(auth.CreateModelArtsNotebookForm{}), repo.NotebookCreate) + }) + + m.Group("/train-job", func() { + m.Get("", reqRepoCloudBrainReader, repo.TrainJobIndex) + m.Group("/:jobid", func() { + m.Get("", reqRepoCloudBrainReader, repo.TrainJobShow) + m.Post("/stop", reqRepoCloudBrainWriter, repo.TrainJobStop) + m.Post("/del", reqRepoCloudBrainWriter, repo.TrainJobDel) + m.Get("/log", reqRepoCloudBrainReader, repo.TrainJobGetLog) + }) + m.Get("/create", reqRepoCloudBrainReader, repo.TrainJobNew) + m.Post("/create", reqRepoCloudBrainWriter, bindIgnErr(auth.CreateModelArtsTrainJobForm{}), repo.TrainJobCreate) + m.Get("/para-config-list", reqRepoCloudBrainReader, repo.TrainJobGetConfigList) + }) }, context.RepoRef()) m.Group("/blockchain", func() { diff --git a/templates/repo/header.tmpl b/templates/repo/header.tmpl index cf3386df2..c290de552 100755 --- a/templates/repo/header.tmpl +++ b/templates/repo/header.tmpl @@ -176,13 +176,13 @@
- +
- +
@@ -209,14 +209,16 @@ $('.ui.radio.checkbox').checkbox(); var repolink = $(".cloudbrain_link").text() + console.log(repolink) $(".ui.positive.right.icon.button").click(function(){ // 声明一个变量来接收以及获取单选框选择的情况 - var checked_radio = $("input[type='radio']:checked").val() + var checked_radio = $("input[name='CloudBrainSelect']:checked").val() + console.log(checked_radio) if(checked_radio=='0'){ window.location.href = repolink+'/cloudbrain' }else if(checked_radio=='1'){ - window.location.href = repolink+'/modelarts' + window.location.href = repolink+'/modelarts/notebook' }else{ return; } diff --git a/templates/repo/modelarts/index.tmpl b/templates/repo/modelarts/index.tmpl index 7820cb109..fc804c393 100755 --- a/templates/repo/modelarts/index.tmpl +++ b/templates/repo/modelarts/index.tmpl @@ -214,6 +214,11 @@ {{if .Permission.CanWrite $.UnitTypeCloudBrain}} {{.i18n.Tr "repo.cloudbrain.new"}} {{end}} + +
+ {{if .Permission.CanWrite $.UnitTypeCloudBrain}} + {{.i18n.Tr "repo.cloudbrain.new"}} {{end}} +

使用鹏城云脑计算资源进行调试,云脑1提供CPU / GPU资源,云脑2提供Ascend NPU资源;调试使用的数据集也需要上传到对应的环境。

diff --git a/templates/repo/modelarts/navbar.tmpl b/templates/repo/modelarts/navbar.tmpl new file mode 100755 index 000000000..91c0675c0 --- /dev/null +++ b/templates/repo/modelarts/navbar.tmpl @@ -0,0 +1,43 @@ + +
+
+ +
+
\ No newline at end of file diff --git a/templates/repo/modelarts/notebook/index.tmpl b/templates/repo/modelarts/notebook/index.tmpl new file mode 100755 index 000000000..cb93c0a32 --- /dev/null +++ b/templates/repo/modelarts/notebook/index.tmpl @@ -0,0 +1,221 @@ +{{template "base/head" .}} + +
+
+
+ {{template "repo/header" .}} +
+
+ {{template "repo/modelarts/navbar" .}} + +
+
+
+

{{.i18n.Tr "repo.modelarts.notebook"}}

+
+
+
+
+ {{if .Permission.CanWrite $.UnitTypeCloudBrain}} + {{.i18n.Tr "repo.cloudbrain.new"}} {{end}} +
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+ + +
+ {{range .Tasks}} +
+
+ + + + +
+ {{.Status}} +
+ + +
+ {{svg "octicon-flame" 16}} {{TimeSinceUnix .CreatedUnix $.Lang}} +
+ + +
+ + + 查看 + + +
+ + +
+
+
+ {{$.CsrfTokenHtml}} + 删除 +
+
+
+ + +
+ +
+ + +
+
+
+ {{$.CsrfTokenHtml}} + 停止 +
+
+
+
+
+ {{end}} {{template "base/paginate" .}} +
+ +
+
+
+ +
+
+
+
+ + +
+ +
+
+{{template "base/footer" .}} + + diff --git a/templates/repo/modelarts/notebook/new.tmpl b/templates/repo/modelarts/notebook/new.tmpl new file mode 100755 index 000000000..82c267a74 --- /dev/null +++ b/templates/repo/modelarts/notebook/new.tmpl @@ -0,0 +1,82 @@ +{{template "base/head" .}} +
+
{{.i18n.Tr "loading"}}
+
+
+ {{template "repo/header" .}} +
+
+ {{template "base/alert" .}} +
+ {{.CsrfTokenHtml}} +

+ {{.i18n.Tr "repo.cloudbrain.new"}} +

+
+ +
+ + +
+ +
+ + +
+ +
+ + +
+
+ + +
+
+ + +
+
+ + +
+
+ + +
+
+ + + {{.i18n.Tr "repo.cloudbrain.cancel"}} +
+
+
+
+
+
+{{template "base/footer" .}} + + \ No newline at end of file diff --git a/templates/repo/modelarts/notebook/show.tmpl b/templates/repo/modelarts/notebook/show.tmpl new file mode 100755 index 000000000..3f914b56d --- /dev/null +++ b/templates/repo/modelarts/notebook/show.tmpl @@ -0,0 +1,122 @@ +{{template "base/head" .}} +
+{{template "repo/header" .}} +
+
+ {{template "base/alert" .}} + +

+ 返回 +

+
+
+ {{with .task}} +

任务名称: {{.JobName}}

+ {{end}} +
+
+

任务结果:

+ {{with .result}} + + + + + + + + + + + + + + + +
状态 {{.Status}}
开始时间 {{.CreateTime}}
最后更新时间 {{.LatestUpdateTime}}
+ {{end}} +
+
+ {{with .result}} + + + + + + + + + + + + + + +
配置信息
开发环境类型 {{.Profile.DeType}}
硬件类型 {{.Profile.FlavorType}}
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
机器规格详情
机器规格 {{.Flavor}}
规格名称 {{.FlavorDetails.Name}}
规格销售状态 {{.FlavorDetails.Status}}
排队个数 {{.FlavorDetails.QueuingNum}}
排到队的剩余时间(秒) {{.FlavorDetails.QueueLeftTime}}
自动停止时间(秒) {{.FlavorDetails.Duration}}
+ + + + + + + + + + + + + + + + + + + + + + + + + + + +
排队信息
实例状态 {{.QueuingInfo.Status}}
实例排队的开始时间 {{.QueuingInfo.BeginTime}}
排到队的剩余时间(秒) {{.QueuingInfo.RemainTime}}
实例排队的预计停止时间 {{.QueuingInfo.EndTime}}
实例在队列中的排位 {{.QueuingInfo.Rank}}
+ {{end}} +
+
+ +
+
+
+{{template "base/footer" .}} diff --git a/templates/repo/modelarts/trainjob/edit_para.tmpl b/templates/repo/modelarts/trainjob/edit_para.tmpl new file mode 100755 index 000000000..a38b2a4f0 --- /dev/null +++ b/templates/repo/modelarts/trainjob/edit_para.tmpl @@ -0,0 +1,239 @@ +{{template "base/head" .}} +
+
{{.i18n.Tr "loading"}}
+
+ +
+ {{template "repo/header" .}} +
+ {{template "base/alert" .}} +

+ {{.i18n.Tr "repo.modelarts.train_job.new"}} +

+
+ +
+ {{.CsrfTokenHtml}} + +

{{.i18n.Tr "repo.modelarts.train_job.basic_info"}}

+
+ + +
+
+ + +
+

{{.i18n.Tr "repo.modelarts.train_job.parameter_setting"}}

+
+ + +
+
+ +
+
+ +
+
+ +
+
+
+
+ + + + + +
+
+
+
+ + +
+ +
+ + {{.i18n.Tr "repo.modelarts.train_job.add_run_parameter"}} + +
+ {{range .para}} +
+
+ +
+
+ +
+ + + + +
+ {{end}} +
+
+ +

{{.i18n.Tr "repo.modelarts.train_job.resource_setting"}}

+
+ + +
+ +
+ + {{range .benchmark_categories}} +
+
+
+
+ +
+
+
train-private-1
+
{{svg "octicon-verified" 16}} 运行中
+
CPU:192 核 2048GiB
+
+
+ {{end}} +
+ +
+ + +
+
+ + +
+
+ + {{.i18n.Tr "repo.cloudbrain.cancel"}} +
+
+
+
+
+{{template "base/footer" .}} + + \ No newline at end of file diff --git a/templates/repo/modelarts/trainjob/index.tmpl b/templates/repo/modelarts/trainjob/index.tmpl new file mode 100755 index 000000000..c6e1b3c3e --- /dev/null +++ b/templates/repo/modelarts/trainjob/index.tmpl @@ -0,0 +1,225 @@ +{{template "base/head" .}} + +
+
+
+ {{template "repo/header" .}} + +
+ {{template "base/alert" .}} +
+ {{template "repo/modelarts/navbar" .}} + +
+
+
+

{{.i18n.Tr "repo.modelarts.train_job"}}

+
+
+
+
+ {{if .CanCreate}} + {{.i18n.Tr "repo.modelarts.train_job.new"}} + {{end}} +
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+ + +
+ {{range .Tasks}} +
+
+ + + + +
+ {{.Status}} +
+ + +
+ {{svg "octicon-flame" 16}} {{TimeSinceUnix .CreatedUnix $.Lang}} +
+ + +
+
+
+ {{$.CsrfTokenHtml}} + 删除 +
+
+
+ + +
+
+
+ {{$.CsrfTokenHtml}} + 停止 +
+
+
+ +
+
+ {{end}} {{template "base/paginate" .}} +
+
+
+
+
+
+
+
+ + +
+ +
+
+{{template "base/footer" .}} + + diff --git a/templates/repo/modelarts/trainjob/new.tmpl b/templates/repo/modelarts/trainjob/new.tmpl new file mode 100755 index 000000000..bad82abb0 --- /dev/null +++ b/templates/repo/modelarts/trainjob/new.tmpl @@ -0,0 +1,407 @@ +{{template "base/head" .}} +
+
{{.i18n.Tr "loading"}}
+
+ +
+ {{template "repo/header" .}} +
+ {{template "base/alert" .}} +

+ {{.i18n.Tr "repo.modelarts.train_job.new"}} +

+
+ +
+ {{.CsrfTokenHtml}} + +

{{.i18n.Tr "repo.modelarts.train_job.basic_info"}}

+
+ + +
+ +
+ + +
+ +
+ + +
+
+ +
+
+ +
+
+ +
+
+
+
+ + + + + +
+
+
+
+ + +
+ +
+ + {{.i18n.Tr "repo.modelarts.train_job.add_run_parameter"}} + +
+
+
+ +

{{.i18n.Tr "repo.modelarts.train_job.resource_setting"}}

+
+ + +
+ +
+ +
+
+
+
+ +
+
+
train-private-1
+
{{svg "octicon-verified" 16}} 运行中
+
CPU:192 核 2048GiB
+
+
+
+ +
+ + +
+
+ + +
+ + +
+ + {{.i18n.Tr "repo.cloudbrain.cancel"}} +
+ + + +
+
+
+
+{{template "base/footer" .}} + + \ No newline at end of file diff --git a/templates/repo/modelarts/trainjob/para_manage.tmpl b/templates/repo/modelarts/trainjob/para_manage.tmpl new file mode 100755 index 000000000..b8ac6c78a --- /dev/null +++ b/templates/repo/modelarts/trainjob/para_manage.tmpl @@ -0,0 +1,154 @@ +{{template "base/head" .}} +
+
+ {{template "repo/header" .}} +
+
+ {{template "repo/modelarts/navbar" .}} + +
+
+
+

{{.i18n.Tr "repo.modelarts.train_job_para_admin"}}

+
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+ + +
+ {{range .Tasks}} +
+
+ + + + +
+ {{.Status}} +
+ + +
+ {{svg "octicon-clock" 16}} {{TimeSinceUnix .CreatedUnix $.Lang}} +
+ + + + + +
+
+
+ {{$.CsrfTokenHtml}} + 删除 +
+
+
+ +
+
+ {{end}} {{template "base/paginate" .}} +
+
+
+
+
+
+
+
+ + +
+ +
+
+{{template "base/footer" .}} + + diff --git a/templates/repo/modelarts/trainjob/show.tmpl b/templates/repo/modelarts/trainjob/show.tmpl new file mode 100755 index 000000000..5616c27be --- /dev/null +++ b/templates/repo/modelarts/trainjob/show.tmpl @@ -0,0 +1,200 @@ +{{template "base/head" .}} +
+{{template "repo/header" .}} +
+
+ {{template "base/alert" .}} +

+
+
+ {{$.i18n.Tr "repo.modelarts.version_manage"}} +
+ +
+

+ +
+ +
+
+ + {{$.i18n.Tr "repo.modelarts.train_job.version"}} +
+
+
+ +
+
+
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
{{.i18n.Tr "repo.modelarts.train_job.basic_info"}}
{{.i18n.Tr "repo.modelarts.train_job.job_name"}} {{.result.JobName}}
{{.i18n.Tr "repo.modelarts.train_job.job_status"}} {{.result.Status}}
{{.i18n.Tr "repo.modelarts.train_job.version"}} {{.result.VersionName}}
{{.i18n.Tr "repo.modelarts.train_job.start_time"}} {{.result.CreateTime}}
{{.i18n.Tr "repo.modelarts.train_job.dura_time"}} {{.result.Duration}}
{{.i18n.Tr "repo.modelarts.train_job.description"}} {{.result.Description}}
+
+
+ + + + + + + + + + + + + + + + + + + + + + +
{{.i18n.Tr "repo.modelarts.train_job.parameter_setting_info"}}
{{.i18n.Tr "repo.modelarts.train_job.AI_driver"}} {{.result.EngineName}} | {{.result.EngineVersion}}
{{.i18n.Tr "repo.modelarts.train_job.start_file"}}{{.result.BootFileUrl}}
{{.i18n.Tr "repo.modelarts.train_job.dataset"}} {{.result.DatasetName}}
{{.i18n.Tr "repo.modelarts.train_job.run_parameter"}} {{.result.Parameter}}
+
+
+ + + + + + + + + + + + + + + + + + +
{{.i18n.Tr "repo.modelarts.train_job.resource_setting_info"}}
{{.i18n.Tr "repo.modelarts.train_job.resource_pool"}} {{.result.PoolName}}
{{.i18n.Tr "repo.modelarts.train_job.amount_of_compute_node"}}{{.result.WorkServerNum}}
{{.i18n.Tr "repo.modelarts.train_job.NAS_mount_path"}} {{.result.NasMountPath}}
+
+
+
+
+ +
+
+ + {{.log_file_name}} + + + +
+
+
+
{{.log.Content}}
+
+ +
+
+
+
+
+
+
+
+ +{{template "base/footer" .}} + + \ No newline at end of file diff --git a/web_src/less/themes/theme-arc-green.less b/web_src/less/themes/theme-arc-green.less old mode 100644 new mode 100755